February 10, 2005
Tower Records Tunes Its Site
By Evan Schuman, eWEEK
Controlling how people listen to a song on a CD is as easy as humming compared with controlling how 70,000 people every day navigate an e-commerce site, executives at Tower Records have discovered.
Like every other e-commerce outfit, Tower's Web team makes a long list of assumptions about how people will interact with the site right before launching a new site capability. Those assumptions are often wrong because, among other reasons, the thinking process of a Web developer/designer doesn't always mesh with the thoughts of the site's visitors.
The gap between developer assumptions and site visitor realities has widened over the last few years, as visitors have become more comfortable with the Web—and therefore less willing to acquiesce to the site's way of doing things—and more empowered with conflict-friendly software such as firewalls, pop-up blockers and non-traditional browsers.
To get more in synch with visitors, Tower brought in a piece of high-end analytical software from TeaLeaf Technology. Although the software cannot see what is on the customer's screen, it can get a lot closer than typical Web analytics software can, according to Kevin Ertell, Tower Record's vice president of operations.
"We're seeing what they see and what they typed in and what they did," he said. "It allows us to replay sessions that people have so that we can almost see what they see."
For example, one of the Tower assumptions was that consumers would fill out online forms sequentially, answering the questions in the order they were asked. But many customers wanted to answer certain later questions first, an approach that confused the site's validation routines.
"People took a path that we didn't anticipate," Ertell said. "The checkout process is fairly linear. But this customer went right to the bottom of the page where the gift cards were and did that first before filling out any other information on the page. That should have been fine" but instead it delivered some "really ugly errors."
Because the error message was not explicit and customer service representatives didn't anticipate the sequence in which customers would complete forms, the errors baffled customer service until they used the TeaLeaf program to search for transactions where that error materialized. Then they played back the customer's input and discovered the pattern.
Tower is the nation's ninth-largest music Web site, seeing about 70,000 unique visitors—and processing about 3,000 orders—every day, Ertell said.
"One of the big things we're trying to do is make sure that we provide the best possible experience for our customers," Ertell said. "But before we start getting into any whiz-bang technology, the most basic thing is to make sure that the things we do offer work correctly. And that hasn't always been the case."
What the TeaLeaf package attempts to do is recreate what the user is seeing. It has limitations, though. It can't know, for example, what font size the customer has chosen, which might make some information appear beyond the viewing area of the screen. It may know what firewall, pop-up blocker or anti-virus programs are installed and have a list of known conflicts with those programs, but it wouldn't definitively know what impact those programs are having on the customer's screen.
"We know the browser engine, and we know what plug-ins are plugged into the browser. For example, you can see that you have the Google pop-up blocker," said Geoff Galat, TeaLeaf's VP of marketing and product strategy. "We don't see what the effects of the blockers are, but we do know that they're there."
That said, the program analyzes a lot of available information and presents a likely view of what the customer is seeing.
The program sniffs all input at the TCP/IP level, examining "all the traffic that comes in and out. We look at everything the browser is submitting to the app and everything the app is serving out to the browser," Galat said.
Not only does the app recreate the viewer experiences, it catalogues and stores them inside its database for later retailer analysis.
"We're indexing all of the information as it comes through in a Google-like fashion so it all becomes searchable later," Galat said. "So, for example, you could say, 'Go into the system and search for everybody who had Google Pop-up toolbar installed or for everybody who is using IE (Internet Explorer) 3 or below or search for everybody who had the following cookie installed or everybody from a certain IP address.' "
The benefit of such analysis is that it allows an e-commerce administrator to quickly determine whether the situation "is a 'one-person' problem or a 'lots of people' problem."
Tower's Ertell said that he finds the database search capability helpful because customers often are inaccurate when they tell customer service what happened, sometimes focusing on irrelevant details and ignoring technologically significant ones.
"A lot of the times the description of the problem is not really the problem, so having a tool that lets us see what really happened helps," Ertell said. "It truly is able to look for certain patterns and pull all the sessions that match those patterns."
TeaLeaf's package sells for between $75,000 and more than $1 million, with an average package selling for about $170,000, Galat said. The price is based on two measurements: the number of Web servers used and the number of collective CPUs ("a relative way to understand traffic"); and the number of people who will be interacting with the data, including customer service, second-tier support, developers and line-of-business managers, he said.
Tower's Ertell said that another lesson that Tower has learned is how people react to home-page promotions. The current site will trumpet a major promotion at the top of the page and three minor ones below that.
"That may seem obvious now, but in the past we gave every promotion equal treatment on the home page instead of having one obvious one," he said, attributing the change to Web analytics.
Many of the changes that e-commerce sites make after reviewing these kinds of Web analytics programs are not radical or even surprising, but merely reflect an improved understanding of how consumers interact with the pages, Galat said.
For example, an insurance site required site visitors to enter a 17-character Vehicle Identication Number (VIN), Galat said, so they offered a straight text field with enough space to enter the letters and numerals. Extensive testing showed that it worked fine. "100 percent of the time, the application went through," he said.
But it provided no guidance about upper-case characters and whether dashes were to be used. The result? "Literally hundreds of customers abandoned the process," he said.
Once discovered, a new form with lower-case instructions and pre-formatted dashes eliminated the problem, he said.
Another example: TeaLeaf had a client that issued credit cards. Three kinds of credit cards were being offered: silver, gold and platinum. The program tried to get certain income groups into different cards, and so when someone was applying for a Silver card and typed in more than $100,000 in income, it would flash a screen suggesting that they apply for Gold with more benefits. When someone was applying for Gold and typed in more than $150,000 in income, it would do the same, this time suggesting platinum.
The application kept crashing and customer service was baffled. Later analysis showed that all of the failures were people who had applied for the Silver card but had reported more than $150K in income.
"This was something that just snuck through the testing process. No one had ever tested a two-level jump" from silver to platinum, Galat said. "There was this logic error and the application didn't know what to do with it." The application decided to just dump the customer back to the beginning of the application "so the customer kept going through an endless loop."